A great AEO audit deliverable tells the client where they stand in AI answers, why, and exactly what to do first, with evidence they can verify and findings their teams can act on without translation. Most weak audits fail on the last part: they show screenshots and scores but leave the client with a pile of observations. This post gives a table of contents, the format for every finding, a scoring method, and a quality checklist to run before delivery.
The table of contents
Use this structure for most audits, scaling the depth of each section to scope.
- Executive summary (one page)
- Scope and method (half a page; full detail in the appendix)
- Visibility baseline by platform and prompt cluster
- Competitor benchmark
- Accuracy and perception review
- Citation and source analysis
- Technical access and rendering findings
- Content gap review
- Prioritized findings
- 90-day roadmap
- Appendices: prompt set, run log, raw results, glossary
Section 1: The executive summary
The executive summary is one page with five parts: the headline position, the three biggest problems, the three biggest opportunities, the recommended first 30 days, and what success looks like at 90 days. Write it last and test it on someone who hasn't seen the audit. See explaining AEO to executives for language that works with leadership.
Sections 3 and 4: Baseline and benchmark
The baseline section shows mention rate, share of voice and citation share, broken down so patterns are visible.
| Cluster | Prompts | ChatGPT | Perplexity | AI Overviews | Gemini | Copilot |
|---|---|---|---|---|---|---|
| Unbranded shortlist | 30 | |||||
| Comparison | 15 | |||||
| Pricing and plans | 10 | |||||
| Brand facts | 15 |
Fill cells with mention rate across repeated runs, not a single run. Then add the competitor benchmark: the same metrics for each competitor, plus which competitor wins each cluster and why. See competitor benchmarking in AI search.
Sections 5 and 6: Accuracy, perception and sources
These two sections explain why the baseline looks the way it does: what AI gets wrong about the brand, and which sources shape its answers.
Accuracy and perception
List every material inaccuracy, with the prompt, platform, what the answer said, what's true, and the likely source. Then summarize perception: the adjectives and "best for" associations assistants use, and any recurring caveats ("expensive," "limited support"). Inaccuracies are often the most persuasive part of the audit, because they are concrete and fixable. See finding and fixing AI hallucinations about your brand.
Citation and source analysis
Show which domains AI cites for the client's prompts, grouped by type: the client's own site, competitors, review platforms, editorial roundups, communities, news, directories and reference sites. For each high-influence source, note whether the client is present, absent or misrepresented. This becomes the PR and partnerships target list. See citation analysis.
Sections 7 and 8: Technical and content
Technical findings cover robots.txt and CDN rules for AI crawlers, rendering, indexing in Google and Bing, structured data and server log evidence of AI fetchers. Use the technical AEO audit checklist as the working list, and report only what you found, not the whole checklist.
The content gap review maps prompt clusters to existing pages: which prompts have a strong answering page, a weak one, or none. Weak pages get restructure recommendations; missing ones become new page briefs. See the AEO content audit.
The format for every finding
Every finding in section 9 uses the same fields. Consistency is what makes an audit actionable.
| Field | Content |
|---|---|
| ID | T-04, C-11, P-02 (technical, content, PR) |
| Finding | One sentence: what's wrong or missing |
| Evidence | Prompt, platform, date, screenshot or log line |
| Impact | Which prompts or metrics it affects, and why it matters |
| Recommendation | Specific action, not "improve content" |
| Owner | Client team or agency |
| Effort | S, M or L |
| Priority | From the scoring below |
Example finding: "T-04. Cloudflare bot rules return 403 to OAI-SearchBot and PerplexityBot on /pricing. Evidence: server logs [dates]. Impact: pricing prompts on ChatGPT and Perplexity cite a third-party site with outdated plans. Recommendation: allow these user agents for public pages. Owner: client web team. Effort: S. Priority: 1."
Scoring, prioritization and the 90-day roadmap
Prioritize findings with a simple score, then turn the ranked list into a dated plan.
Scoring
Score each finding on impact (1 to 3) and effort (1 to 3, where 1 is easy), then rank by impact minus effort, breaking ties by the importance of the prompt cluster. Keep the scoring simple enough that the client can re-score findings when their circumstances change.
| Priority | Rule of thumb |
|---|---|
| 1 | High impact, low effort: access blocks, wrong facts on owned pages |
| 2 | High impact, higher effort: new comparison pages, review program |
| 3 | Moderate impact: schema, secondary content updates |
| 4 | Monitor or revisit next quarter |
The 90-day roadmap
Turn priorities into a month-by-month plan with owners, so the audit ends with a decision rather than a reading assignment.
- Days 1–30: all priority 1 findings; prompt set locked; tracking running.
- Days 31–60: the first priority 2 content and source work; re-run baseline on affected clusters.
- Days 61–90: remaining priority 2 items; first progress report against the baseline.
Include what the client should expect to see by day 90 and what will take longer, such as third-party coverage.
Appendices and the pre-delivery QA check
The appendices make the audit reproducible: the full prompt set with clusters, the run log (platform, date, location settings, number of runs), raw results as a spreadsheet, and definitions of each metric. Without them, the next baseline can't be compared to this one.
Before delivery, run this checklist:
- Every finding has evidence, owner, effort and priority
- Baseline numbers come from repeated runs, with method documented
- No finding recommends something the client can't do (check their CMS and team)
- Inaccuracies are verified against the client's own facts, not assumed
- Screenshots are dated and labeled with platform and prompt
- Executive summary fits on one page and stands alone
- Roadmap has dates and named owners
- Raw data file is attached
Example: a structure for a 60-prompt audit
This example is illustrative. A 60-prompt audit for a regional bank across five platforms produces 27 findings: 8 technical, 11 content, 6 PR and sources, and 2 governance. Six are priority 1, including a blocked AI crawler on rate pages and two outdated fee descriptions. The report is 18 pages plus a spreadsheet; the executive summary and roadmap are presented in a 45-minute session, and the findings spreadsheet becomes the first month's backlog. The scope that led to it came from the AEO proposal template.
The bottom line
A great AEO audit is reproducible, evidence-based and built to be acted on. Use a consistent structure, a single format for every finding, simple scoring and a dated roadmap with owners, document the method in appendices, and present it live. The audit's job isn't to impress; it's to make the next 90 days obvious.
